--- title: "AI Reprices SaaS: Why Seat Pricing Breaks and Proprietary Data Wins" description: "Sami Miettinen's Arctic15 2026 keynote: as agentic work converts labour budgets into token demand, inference cost moves into COGS and reprices every SaaS metric. SaaS does not disappear — seat pricing does." role: "keynote" event: "Arctic15" date: 2026-06-11 location: "Cable Factory, Helsinki" language: "en" partners: ["DLA Piper","Fenerum"] slides: https://www.neuvottelija.com/events/ai-reprices-saas-arctic15.pdf canonical: https://www.neuvottelija.com/events/ai-reprices-saas-arctic15/ --- # AI Reprices SaaS: Why Seat Pricing Breaks and Proprietary Data Wins *Keynote delivered at **Arctic15** in Helsinki on 11 June 2026, in a session hosted with **DLA Piper** and **Fenerum**. The written summary below follows the delivered deck and its speaker notes; the [full slide deck is available as a PDF](/events/ai-reprices-saas-arctic15.pdf).* ## The thesis **SaaS does not disappear. Seat pricing does.** That one line runs through the whole talk. The fear in the room — that AI ends software-as-a-service — is misplaced. What AI ends is the *per-seat* business model. Software is moving from selling access to selling work, and the economics of that shift reprice how SaaS companies are valued. The argument has four beats: the macro pool AI draws from, the unit economics it changes, the valuation evidence, and the strategic implication. ## 1. Agentic work converts labour budgets into token demand The global labour market in 2024 was roughly **€44 trillion** — €43,866 billion worldwide, of which Finland is about €134 billion, or 0.3%. That entire pool is what agentic work draws from. As tasks convert into agent runs, labour spend reappears as demand for inference tokens. Goldman Sachs Research (May 2026) estimates that the monthly token volume in agentic applications will grow **24× by 2030**, reaching on the order of 120 quadrillion tokens per month, with enterprise use leapfrogging consumer. The punchline: **inference is no longer an IT expense — it becomes a labour-substitution cost.** Sizing why even small shifts are enormous: replacing roughly 1% of labour is on the order of €440 billion of AI value; a 10% shift is about €4,400 billion — some 33× Finland's entire labour spend. ## 2. The SaaS scorecard gets repriced by inference AI doesn't add a metric — it changes the economics behind the ones you already run. When inference cost lands inside COGS, it reprices margins, growth and valuation at once. The traditional SaaS KPIs — **Gross Margin, EBITDA, Rule of 40, CAC Payback, ARR/FTE** — don't disappear, but they are now driven by a cost that used to be near-zero. Alongside them a set of AI-era metrics becomes essential: - **Inference COGS %** — how much of revenue is eaten by model calls - **Tokens per active user** — the real unit of consumption - **Cost per workflow** and **cost per outcome** — pricing tied to work done - **Churn**, read against how much AI capability is actually attached and used ## 3. Vertical SaaS still defends its premium The valuation evidence: at Q1 2026, **vertical software still trades roughly 79% above horizontal** on median EV / LTM revenue — about **3.4× versus 1.9×**. On forward Rule of 40 ('26E), the vertical median is around **36%** versus **28%** for horizontal. (Source: Translink Corporate Finance quarterly SaaS valuation review, Q1 2026.) The gap holds for structural reasons: domain depth and system-of-record lock-in, higher switching costs and stickier data, and the fact that **AI commoditises generic, horizontal workflows first — not systems of record.** Multiples compressed across the board into 2026, but the vertical/horizontal spread persisted. ## 4. From seats to outcomes The strategic implication reduces to three moves: 1. **Seats decline.** Agents need data access, not user seats. Per-seat licensing stops scaling with the value delivered. 2. **Usage and outcomes rise.** Customers pay for completed workflows — metered via MCP / API — rather than headcount. 3. **Proprietary data wins.** As inference commoditises the application tier, the data layer underneath captures the margin. The Nordic proof point is **ICEYE**: a €10B+ valuation built on a synthetic-aperture-radar sensor-data hypervertical with outcome-driven economics. (Round context: General Atlantic / ICEYE, 9 June 2026 — a €1B+ round including a €450M primary Series F. The €10B+ is the valuation, not the raise. The "Service-as-a-Software" framing is Translink's.) ## The takeaway - SaaS is **not** disappearing. - Software is moving from **selling access to selling work.** - The winners will **own the data and workflows** agents need to do that work. - In shorthand: **SaaS+ = Hypervertical-as-a-Software.** **So what?** For **founders**: know your inference cost per workflow, price outcomes rather than seats, and protect proprietary data. For **investors**: calculate post-inference gross margin, seek system-of-record moats, value data ownership, and treat high churn as a red flag. Slides: https://www.neuvottelija.com/events/ai-reprices-saas-arctic15.pdf --- Cite as: Sami Miettinen, Neuvottelija — AI Reprices SaaS: Why Seat Pricing Breaks and Proprietary Data Wins, Arctic15, 2026-06-11, https://www.neuvottelija.com/events/ai-reprices-saas-arctic15/.